CN111626172A - Device and method for accelerating analysis of similarity of human face features - Google Patents
Device and method for accelerating analysis of similarity of human face features Download PDFInfo
- Publication number
- CN111626172A CN111626172A CN202010435467.4A CN202010435467A CN111626172A CN 111626172 A CN111626172 A CN 111626172A CN 202010435467 A CN202010435467 A CN 202010435467A CN 111626172 A CN111626172 A CN 111626172A
- Authority
- CN
- China
- Prior art keywords
- neural network
- analysis
- intermediate node
- pair
- network model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/168—Feature extraction; Face representation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/172—Classification, e.g. identification
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Abstract
The invention discloses a method for accelerating human face feature similarity analysis, which aims at carrying out neural network model operation on an analysis pair to obtain w layers of intermediate node features and tail end node features; judging whether the feature similarity of the intermediate node features of the layer is greater than a corresponding intermediate threshold value or not according to the intermediate node features obtained each time, if so, outputting the conclusion that the two human face pictures in the analysis pair are the same person, and stopping the operation of the neural network model of the analysis pair; and if the feature similarity of the w-layer intermediate node features is smaller than the corresponding intermediate threshold, judging whether the feature similarity of the tail end node features is larger than the tail end threshold, and outputting a judgment conclusion. The device and the method for accelerating the analysis of the similarity of the human face features, provided by the invention, utilize the intermediate node features of the neural network model to carry out the analysis of the similarity of the human face features, so that the analysis efficiency and the accuracy of an analysis result can be improved.
Description
Technical Field
The invention relates to face recognition, in particular to a device and a method for accelerating face feature similarity analysis.
Background
Since video monitoring is rapidly popularized, a rapid identification technology under a remote and user-uncoordinated state is urgently needed for numerous video monitoring applications, so that the identity of personnel can be rapidly confirmed remotely, and intelligent early warning can be realized. The face recognition technology is undoubtedly the best choice, and the rapid face detection technology can be adopted to search the face from the monitoring video image in real time and compare the face with the face database in real time, so as to realize rapid identity recognition.
In the face recognition process, a deep neural network is mostly adopted, however, the deep neural network usually has higher delay due to large calculation amount, and in embedded devices such as mobile phones, a shallower neural network is usually adopted to achieve lower delay, but the corresponding accuracy rate is also reduced.
Face recognition typically involves 1:1 and 1: X tests. The 1:1 test indicates whether two face images are the same person, and the 1: X test indicates whether one face image exists in a database formed by X face images. The training of the neural network on the face picture generally firstly carries out dummy variable coding on the face picture, and then trains data by utilizing a softmax-cross entropy loss function. The last few layers of the neural network are typically, in turn, a feature layer (possibly in the form of a fully-connected layer, a global pooling layer, etc.), a fully-connected layer, a softmax activation layer, a cross-entropy loss function. For 1: the application of the 1 test is generally to calculate the euclidean distance between the features of the feature layer and the features in the database, and determine whether the features are the same person. 1: the N test may be the above method lasting N times.
The output of any layer of the convolutional neural network can be regarded as a set of features, and the deeper the layer number, the finer the feature processing. When a model scores high on either the validation set or the test set, the extraction of features on the data set by the representative model is almost without loss of valid information. However, the valid information itself cannot be generated by null, so the amount of valid information contained in each layer output of the convolutional neural network should be no less than that of the final output. Because the output of the end of the neural network can judge whether two persons are the same person, the characteristics generated by the intermediate node of the neural network can also distinguish whether two persons are the same person in a certain judging domain, and if the characteristics of the intermediate node can be adopted for judging, the accuracy and the efficiency of the judgment can be greatly increased.
Disclosure of Invention
The invention aims to provide a device and a method for accelerating human face feature similarity analysis, which can improve the analysis efficiency and the accuracy of an analysis result by utilizing the intermediate node features of a neural network model to carry out the human face feature similarity analysis.
In order to achieve the purpose, the invention adopts the following technical scheme: a method for accelerating human face feature similarity analysis comprises the following steps:
s01: training the neural network model by adopting a training set to obtain a trained neural network model; the training set comprises M training pairs, wherein each training pair comprises two human face pictures and a label for judging whether the human face pictures are the same person or not; m is an integer greater than 0;
s02: carrying out neural network model operation on the test set to obtain w-layer intermediate node characteristics; determining an intermediate threshold corresponding to the characteristics of the intermediate nodes on each layer; the test set comprises N test pairs, and the test pairs comprise two human face pictures and a label for judging whether the human face pictures are the same person or not; n and w are integers which are larger than 0, and w is smaller than the total number of output layers of the neural network model;
s03: performing neural network model operation aiming at the analysis pair to obtain w layers of intermediate node characteristics and tail end node characteristics; the analysis pair comprises two human face pictures; judging whether the feature similarity of the intermediate node features of the layer is greater than a corresponding intermediate threshold value or not according to the intermediate node features obtained each time, if so, outputting the conclusion that the two human face pictures in the analysis pair are the same person, and stopping the operation of the neural network model of the analysis pair; if the feature similarity of the w-layer intermediate node features is smaller than the corresponding intermediate threshold, the step S04 is entered;
s04: and judging whether the feature similarity of the terminal node features is greater than a terminal threshold, if so, outputting the conclusion that the two face pictures in the analysis pair are the same person, and if not, outputting the conclusion that the two face pictures in the analysis pair are not the same person.
Further, the intermediate threshold and the end threshold are euclidean distance thresholds or mahalanobis distance thresholds.
Further, the specific method for determining the ith layer intermediate threshold in step S02 is as follows:
s021: respectively acquiring the characteristics of the intermediate nodes of the ith layer of the two pictures in each test pair; i is less than the total number of output layers of the neural network model;
s022: respectively calculating the Mahalanobis distance of the characteristics of the intermediate nodes of the ith layer of the two pictures in each test pair;
s023: and determining the mahalanobis distance threshold according to the mahalanobis distance of each test pair and whether the test pair is a label of the same person.
Further, the method for calculating mahalanobis distance in step S022 includes:
carrying out neural network model operation on the kth test pair in the test set to obtain N of the characteristics of the intermediate node of the ith layer of the two picturesiEach is characterized by Ai,k＝[ai,1,k，ai,2,k，…，ai,Ni,k]、Bi,k＝[bi,1,k，bi,2,k，…，bi,Ni,k]；
Covariance coefficient array C for computing i-th layer intermediate node characteristicsiWherein the covariance coefficient array CiThe coefficient of the m row and the n column is the variance of the m characteristic of the first picture and the n characteristic of the second picture in the test pair;
using covariance coefficient array CiThe inverse matrix of (a) calculates the mahalanobis distance of the test to the i-th level intermediate node feature.
Further, the step S03 analyzes one of the pictures obtained by the sensor; another picture is obtained from the database; the database comprises X personal face samples; x is an integer greater than 0.
Further, the pictures obtained by the sensor and the X personal face samples form X analysis pairs, and the steps S03-S04(X-1) are repeated for confirming the pictures obtained by the sensor.
A device for accelerating human face feature similarity analysis comprises a neural network acceleration unit, a storage unit and a main processing unit, wherein the neural network acceleration unit performs neural network model operation on an analysis pair to obtain w layers of intermediate node features and tail end node features, and stores the w layers of intermediate node features and tail end node features in the storage unit; aiming at the intermediate node characteristics acquired each time, the main processing unit judges whether the characteristic similarity of the intermediate node characteristics at the layer is greater than a corresponding intermediate threshold value or not; if the number of the face images is larger than or equal to the corresponding middle threshold value, outputting the conclusion that the two face images in the analysis pair are the same person, and stopping the operation of the neural network model of the analysis pair; if the feature similarity of the w-layer intermediate node features is smaller than the corresponding intermediate threshold, the main processing unit judges whether the feature similarity of the end node features is larger than an end threshold, if so, the main processing unit outputs the conclusion that the two face pictures in the analysis pair are the same person, and if not, the main processing unit outputs the conclusion that the two face pictures in the analysis pair are not the same person; wherein w is an integer greater than 0 and is less than the total number of output layers of the neural network model.
Further, when the neural network acceleration unit outputs the ith intermediate node feature, the neural network acceleration unit sends an interrupt signal to the main processing unit, and after receiving the interrupt signal, the main processing unit controls the ith intermediate node feature to be stored in the storage unit and then sends an interrupt reset signal to the neural network acceleration unit; i is less than the total number of output layers of the neural network model.
Further, the neural network acceleration unit outputs the ith layer of intermediate node characteristics, transmits the ith layer of intermediate node characteristics to the storage unit in a direct memory access mode, and sends an interrupt signal to the main processing unit after transmission is completed; i is less than the total number of output layers of the neural network model.
The invention has the following beneficial effects: after the intermediate threshold value is determined, the face feature similarity analysis is carried out by using the intermediate node features of the neural network model, so that the analysis efficiency and the accuracy of an analysis result can be improved; and the distance between the middle node and the tail end keeps a certain network depth, so that the acceleration effect is obvious.
Drawings
FIG. 1 is a flow chart of a method for accelerating analysis of similarity of human face features according to the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, embodiments of the present invention are described in detail below with reference to the accompanying drawings.
As shown in fig. 1, the method for accelerating the analysis of the similarity of the human face features provided by the invention comprises the following steps:
s01: training the neural network model by adopting a training set to obtain a trained neural network model; the training set comprises M training pairs, and the training pairs comprise two face pictures and a label for judging whether the face pictures are the same person or not; m is an integer greater than 0. Through the training of a plurality of training pairs, a neural network model with accurate parameters can be obtained, and the trained neural network model is adopted in the subsequent calculation process.
S02: carrying out neural network model operation on the test set to obtain w-layer intermediate node characteristics; determining an intermediate threshold corresponding to the characteristics of the intermediate nodes on each layer; the test set comprises N test pairs, and the test pairs comprise two human face pictures and a label for judging whether the human face pictures are the same person or not; n and w are integers which are larger than 0, and w is smaller than the total output layer number of the neural network model. The test set and the training set are basically the same in composition and are only functionally distinguished, the training set is used for training the neural network model, and the test set is used for determining the intermediate threshold. The intermediate node of the ith layer can be any node before the end node, a plurality of intermediate nodes can be arranged, and the distance between the first arranged intermediate node and the end node is larger, so that the acceleration effect is obvious.
Judging whether the two face pictures in the test pair or the training pair are the same person or not by adopting the Euclidean distance or the Mahalanobis distance; when the two pictures are the same person, the mahalanobis distance between the intermediate node features and the mahalanobis distance between the end node features are smaller, and when the two pictures are not the same person, the mahalanobis distance between the intermediate node features and the mahalanobis distance between the end node features are larger. Therefore, the intermediate threshold and the terminal threshold can be Euclidean distance thresholds or Mahalanobis distance thresholds.
Because the characteristics are subjected to screening of the pooling layer and the activation layer and combination of the convolutional layer and the full-connection layer in the propagation of the characteristics in the neural network, the characteristic distance of the middle node of the neural network can be approximated by the mahalanobis distance, and the coefficient of the mahalanobis distance can be solved by a full probability formula. The scheme of judging whether the people are the same person or not by using the Euclidean distance can restrict the index of the accuracy rate in a threshold value mode. The smaller the threshold value is set, the higher the accuracy rate is, and since the mahalanobis distance judgment of the intermediate node is an approximate solution, the setting of the mahalanobis distance threshold value should be set with reference to the same accuracy rate under the euclidean distance. The following description takes the intermediate threshold as the mahalanobis distance threshold as an example, and when the intermediate threshold is the euclidean distance or other characteristic values, the method for determining the intermediate threshold is similar; the method for determining the mahalanobis distance threshold specifically comprises the following steps:
s021: respectively acquiring the characteristics of the intermediate nodes of the ith layer of the two pictures in each test pair; i is less than the total number of output layers of the neural network model;
s022: respectively calculating the Mahalanobis distance of the characteristic features of the intermediate node at the ith layer of the two pictures in each test pair; carrying out neural network model operation on the kth test pair in the test set to obtain N of the characteristics of the intermediate node of the ith layer of the two picturesiEach is characterized by Ai,k＝[ai,1,k，ai,2,k，…，ai,Ni,k]、Bi,k＝[bi,1,k，bi,2,k，…，bi,Ni,k]；
Covariance coefficient array C for computing i-th layer intermediate node characteristicsiWherein the covariance coefficient array CiThe coefficient of the m row and the n column is the variance of the m characteristic of the first picture and the n characteristic of the second picture in the test pair;
using covariance coefficient array CiThe mahalanobis distance of the test to the characteristics of the intermediate node at the ith layer is calculated by the inverse matrix of (1), and the specific calculation method is described in the prior art and is not described in detail here.
S023: and determining the mahalanobis distance threshold according to the mahalanobis distance of each test pair and whether the test pair is the label of the same person, and considering the allowable misrecognition rate.
The terminal threshold value can also be determined by adopting the method, and the terminal node characteristics can be treated as special intermediate node characteristics; the termination threshold may also be determined in any manner known in the art.
S03: performing neural network model operation aiming at the analysis pair to obtain w layers of intermediate node characteristics and tail end node characteristics; the analysis pair comprises two face pictures. Wherein, when the present invention is applied to 1:1, analyzing the two pictures in the pair respectively to be the picture obtained by the sensor and the picture to be compared. When the present invention is applied to 1: when the mode of X is determined, one picture of the analysis pair is obtained through a sensor; another picture is obtained from the database; the database comprises X personal face samples; x is an integer greater than 0. And (4) forming X analysis pairs by the pictures obtained by the sensor and X personal face samples, repeating the steps S03-S04X-1 times, and confirming the pictures obtained by the sensor. Note that: the purpose of steps S01-S02 is to obtain a neural network model and an intermediate threshold, and one of the pictures obtained by the sensor in the analysis pair in this step is the picture to be determined by the present invention, so as to determine whether the picture is the same person as the X person face sample in the database.
Judging whether the feature similarity of the intermediate node features of the layer is greater than a corresponding intermediate threshold value or not according to the intermediate node features obtained each time, if so, outputting a conclusion that the two human face pictures in the analysis pair are the same person, and stopping the operation of a neural network model of the analysis pair; if the feature similarity of the w-layer intermediate node features is smaller than the corresponding intermediate threshold, the process proceeds to step S04.
The output of the intermediate node characteristics and the judgment of the characteristic similarity of the intermediate node characteristics are two separate processes, and a neural network accelerating unit performs neural network model operation on an analysis pair to obtain an intermediate node and a tail end node; and the main processing unit judges the feature similarity of the intermediate node features. It is worth to be noted that, in the present invention, after the neural network model outputs the i-th layer intermediate node feature, the operation is continued, and the next intermediate node feature or the end node feature is output, while the neural network model continues to perform the operation, the main processing unit determines the feature similarity for the output intermediate node feature, and if the feature similarity of the layer intermediate node feature is greater than or equal to the corresponding intermediate threshold, the conclusion that two face pictures in the analysis pair are the same person is output, and at the same time, the operation for the neural network model in the analysis pair is stopped, that is, the entire analysis process is ended.
The innovation of the invention is that the characteristics of the intermediate node are output, the face similarity is judged by using the characteristics of the intermediate node, and if the characteristics of the intermediate node and the characteristics of the end node are judged to be the same person, the characteristics of the end node can not be judged, and a conclusion is directly output; if all the intermediate nodes judge that the two nodes are not the same person, at the moment, the judgment of the end node is continued. In this determination mode, the determination efficiency can be accelerated.
S04: judging whether the feature similarity of the terminal node features is greater than a terminal threshold, and if the feature similarity of the terminal node features is greater than or equal to the terminal threshold, outputting a conclusion that the two face pictures in the pair are the same person; and if the number of the face pictures is smaller than the terminal threshold value, outputting a conclusion that the two face pictures in the analysis pair are not the same person.
The invention provides a device for accelerating human face feature similarity analysis, which comprises a neural network acceleration unit, a storage unit and a main processing unit, wherein the neural network acceleration unit performs neural network model operation on an analysis pair to obtain w layers of intermediate node features and tail end node features, and stores the w layers of intermediate node features and tail end node features in the storage unit; aiming at the intermediate node characteristics acquired each time, the main processing unit judges whether the characteristic similarity of the intermediate node characteristics of the layer is greater than the corresponding intermediate threshold value or not according to the intermediate node characteristics stored in the storage unit and the corresponding intermediate threshold value; if the number of the face images is larger than or equal to the corresponding middle threshold value, outputting a conclusion that the two face images in the analysis pair are the same person, and stopping the operation of the neural network model of the analysis pair; if the feature similarity of the w-layer intermediate node features is smaller than the corresponding intermediate threshold, the main processing unit judges whether the feature similarity of the end node features is larger than the end threshold according to the end node features and the corresponding end threshold stored in the storage unit, if so, the main processing unit outputs a conclusion that the two face pictures in the analysis pair are the same person, and if not, the main processing unit outputs a conclusion that the two face pictures in the analysis pair are not the same person; wherein w is an integer greater than 0 and is less than the total number of output layers of the neural network model. The method outputs the characteristics of the intermediate nodes, judges the face similarity by using the characteristics of the intermediate nodes, does not judge the characteristics of the subsequent intermediate nodes and the terminal nodes if one of the characteristics of the intermediate nodes is judged to be the same person, and outputs a conclusion; in this determination mode, the determination efficiency can be accelerated. If all the intermediate node characteristics judge that the two are not the same person, in order to ensure the judging accuracy, the terminal node characteristics can be adopted again for judging, and a conclusion is output.
The invention increases the output and judgment of the intermediate node characteristics, so the arrangement of the storage unit in the hardware has a place different from the prior art, the neural network accelerating unit adopts the neural network model to output the intermediate node characteristics and the end node characteristics, and the output of the intermediate node of the neural network model is premised on that the further calculation of the neural network is not interfered. Specifically, the storage unit may be set in the following two ways:
firstly, when the neural network acceleration unit outputs the ith layer of intermediate node characteristics, the intermediate node characteristics should be stored in an effective storage area and not covered by a subsequent result; at the moment, the neural network acceleration unit sends an interrupt signal to the main processing unit, after the main processing unit receives the interrupt signal, the i-th layer intermediate node feature is controlled to be stored in the storage unit, then an interrupt reset signal is sent to the neural network acceleration unit, and then the main processing unit judges the intermediate node feature. In the invention, the number of times that the neural network acceleration unit sends the interrupt signal to the main processing unit is determined by a code, but the upper limit of the number of times that the neural network acceleration unit sends the intermediate interrupt signal to the main processing unit is limited by the size of a small storage area and the sizes of intermediate node characteristics and end node characteristics under the non-coverage storage form of the storage unit. After the main processing unit finishes judging the characteristics of the end nodes, the characteristics of the middle nodes in the storage interval and the storage data corresponding to the characteristics of the end nodes can be cleaned. Note that the neural network accelerating unit may output the intermediate node feature more than once.
Secondly, when the neural network acceleration unit outputs the intermediate node characteristics of the ith layer, the intermediate node characteristics are transmitted to the storage unit through a Direct Memory Access (DMA), and the writing of the area is forbidden before the DMA transmission is completed; and after the DMA transmission is finished, the neural network acceleration unit sends an interrupt signal to the main processing unit, and the main processing unit judges the characteristics of the intermediate node in the storage unit. And when the main processing unit judges that writing is finished but no effective storage area can be written, the neural network acceleration unit is suspended until writing is started again when the effective storage area exists in the storage unit.
After the intermediate threshold value is determined, the face feature similarity analysis is carried out by using the intermediate node features of the neural network model, so that the analysis efficiency and the accuracy of an analysis result can be improved; and the distance between the middle node and the tail end keeps a certain network depth, so that the acceleration effect is obvious.
The above description is only a preferred embodiment of the present invention, and the embodiment is not intended to limit the scope of the present invention, so that all equivalent structural changes made by using the contents of the specification and the drawings of the present invention should be included in the scope of the appended claims.
Claims (9)
1. A method for accelerating human face feature similarity analysis is characterized by comprising the following steps:
s01: training the neural network model by adopting a training set to obtain a trained neural network model; the training set comprises M training pairs, wherein each training pair comprises two human face pictures and a label for judging whether the human face pictures are the same person or not; m is an integer greater than 0;
s02: carrying out neural network model operation on the test set to obtain w-layer intermediate node characteristics; determining an intermediate threshold corresponding to the characteristics of the intermediate nodes on each layer; the test set comprises N test pairs, and the test pairs comprise two human face pictures and a label for judging whether the human face pictures are the same person or not; n and w are integers which are larger than 0, and w is smaller than the total number of output layers of the neural network model;
s03: performing neural network model operation aiming at the analysis pair to obtain w layers of intermediate node characteristics and tail end node characteristics; the analysis pair comprises two human face pictures; judging whether the feature similarity of the intermediate node features of the layer is greater than a corresponding intermediate threshold value or not according to the intermediate node features obtained each time, if so, outputting the conclusion that the two human face pictures in the analysis pair are the same person, and stopping the operation of the neural network model of the analysis pair; if the feature similarity of the w-layer intermediate node features is smaller than the corresponding intermediate threshold, the step S04 is entered;
s04: and judging whether the feature similarity of the terminal node features is greater than a terminal threshold, if so, outputting the conclusion that the two face pictures in the analysis pair are the same person, and if not, outputting the conclusion that the two face pictures in the analysis pair are not the same person.
2. The method of claim 1, wherein the intermediate threshold and the end threshold are Euclidean distance thresholds or Mahalanobis distance thresholds.
3. The method for accelerating analysis of similarity of human face features according to claim 2, wherein the specific method for determining the intermediate threshold value at the i-th layer in the step S02 is as follows:
s021: respectively acquiring the characteristics of the intermediate nodes of the ith layer of the two pictures in each test pair; i is less than the total number of output layers of the neural network model;
s022: respectively calculating the Mahalanobis distance of the characteristics of the intermediate nodes of the ith layer of the two pictures in each test pair;
s023: and determining the mahalanobis distance threshold according to the mahalanobis distance of each test pair and whether the test pair is a label of the same person.
4. The method for accelerating similarity analysis of human face features according to claim 3, wherein the method for calculating mahalanobis distance in step S022 comprises:
carrying out neural network model operation on the kth test pair in the test set to obtain N of the characteristics of the intermediate node of the ith layer of the two picturesiEach is characterized by Ai,k＝[ai,1,k，ai,2,k，…，ai,Ni,k]、Bi,k＝[bi,1,k，bi,2,k，…，bi,Ni,k]；
Covariance coefficient array C for computing i-th layer intermediate node characteristicsiWherein the covariance coefficient array CiThe coefficient of the m row and the n column is the variance of the m characteristic of the first picture and the n characteristic of the second picture in the test pair;
using covariance coefficient array CiThe inverse matrix of (a) calculates the mahalanobis distance of the test to the i-th level intermediate node feature.
5. The method for accelerating analysis of similarity of human face features according to claim 1, wherein the analysis of step S03 is performed on one of the pictures obtained by a sensor; another picture is obtained from the database; the database comprises X personal face samples; x is an integer greater than 0.
6. The method of claim 5, wherein the pictures obtained by the sensor are combined with X individual face samples to form X analysis pairs, and the steps S03-S04(X-1) are repeated to confirm the pictures obtained by the sensor.
7. A device for accelerating human face feature similarity analysis is characterized by comprising a neural network acceleration unit, a storage unit and a main processing unit, wherein the neural network acceleration unit performs neural network model operation on an analysis pair to obtain w layers of intermediate node features and tail end node features, and stores the w layers of intermediate node features and tail end node features in the storage unit; aiming at the intermediate node characteristics acquired each time, the main processing unit judges whether the characteristic similarity of the intermediate node characteristics at the layer is greater than a corresponding intermediate threshold value or not; if the number of the face images is larger than or equal to the corresponding middle threshold value, outputting the conclusion that the two face images in the analysis pair are the same person, and stopping the operation of the neural network model of the analysis pair; if the feature similarity of the w-layer intermediate node features is smaller than the corresponding intermediate threshold, the main processing unit judges whether the feature similarity of the end node features is larger than an end threshold, if so, the main processing unit outputs the conclusion that the two face pictures in the analysis pair are the same person, and if not, the main processing unit outputs the conclusion that the two face pictures in the analysis pair are not the same person; wherein w is an integer greater than 0 and is less than the total number of output layers of the neural network model.
8. The apparatus of claim 7, wherein when the neural network acceleration unit outputs an i-th layer intermediate node feature, the neural network acceleration unit sends an interrupt signal to the main processing unit, and after receiving the interrupt signal, the main processing unit controls the i-th layer intermediate node feature to be stored in the storage unit and then sends an interrupt reset signal to the neural network acceleration unit; i is less than the total number of output layers of the neural network model.
9. The device for accelerating analysis of similarity of human face features according to claim 7, wherein the neural network acceleration unit outputs the i-th layer intermediate node feature and transmits the i-th layer intermediate node feature to the storage unit in a direct memory access manner, and when the transmission is completed, the neural network acceleration unit sends an interrupt signal to the main processing unit; i is less than the total number of output layers of the neural network model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010435467.4A CN111626172B (en) | 2020-05-21 | 2020-05-21 | Device and method for accelerating analysis of similarity of facial features |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010435467.4A CN111626172B (en) | 2020-05-21 | 2020-05-21 | Device and method for accelerating analysis of similarity of facial features |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111626172A true CN111626172A (en) | 2020-09-04 |
| CN111626172B CN111626172B (en) | 2023-09-08 |
Family
ID=72271042
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010435467.4A Active CN111626172B (en) | 2020-05-21 | 2020-05-21 | Device and method for accelerating analysis of similarity of facial features |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111626172B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106529448A (en) * | 2016-10-27 | 2017-03-22 | 四川长虹电器股份有限公司 | Method for performing multi-visual-angle face detection by means of integral channel features |
| CN107423690A (en) * | 2017-06-26 | 2017-12-01 | 广东工业大学 | A kind of face identification method and device |
| CN108564049A (en) * | 2018-04-22 | 2018-09-21 | 北京工业大学 | A kind of fast face detection recognition method based on deep learning |
-
2020
- 2020-05-21 CN CN202010435467.4A patent/CN111626172B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106529448A (en) * | 2016-10-27 | 2017-03-22 | 四川长虹电器股份有限公司 | Method for performing multi-visual-angle face detection by means of integral channel features |
| CN107423690A (en) * | 2017-06-26 | 2017-12-01 | 广东工业大学 | A kind of face identification method and device |
| CN108564049A (en) * | 2018-04-22 | 2018-09-21 | 北京工业大学 | A kind of fast face detection recognition method based on deep learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111626172B (en) | 2023-09-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106899968B (en) | Active non-contact identity authentication method based on WiFi channel state information | |
| CN110427807B (en) | Time sequence event action detection method | |
| CN109882019B (en) | Automobile electric tail door opening method based on target detection and motion recognition | |
| Bigun et al. | Multimodal biometric authentication using quality signals in mobile communications | |
| CN107967695B (en) | A kind of moving target detecting method based on depth light stream and morphological method | |
| CN108230291B (en) | Object recognition system training method, object recognition method, device and electronic equipment | |
| CN109190470B (en) | Pedestrian re-identification method and device | |
| CN110378219B (en) | Living body detection method, living body detection device, electronic equipment and readable storage medium | |
| CN111401171B (en) | Face image recognition method and device, electronic equipment and storage medium | |
| CN107784215B (en) | Audio unit based on intelligent terminal carries out the user authen method and system of labiomaney | |
| CN113886792A (en) | Application method and system of print control instrument combining voiceprint recognition and face recognition | |
| CN105844204B (en) | Human behavior recognition method and device | |
| CN112183449A (en) | Driver identity verification method and device, electronic equipment and storage medium | |
| CN113343198B (en) | Video-based random gesture authentication method and system | |
| CN110175578A (en) | Micro- expression recognition method based on depth forest applied to criminal investigation | |
| CN113785304A (en) | Face recognition method and device | |
| CN111626172B (en) | Device and method for accelerating analysis of similarity of facial features | |
| US10803875B2 (en) | Speaker recognition system and method of using the same | |
| Bigun et al. | Combining biometric evidence for person authentication | |
| CN114626412B (en) | Multi-class target identification method and system for unattended sensor system | |
| CN115862634A (en) | Voiceprint recognition method and embedded device | |
| CN110956098B (en) | Image processing method and related equipment | |
| CN113887357B (en) | Face representation attack detection method, system, device and medium | |
| CN112669836B (en) | Command recognition method and device and computer readable storage medium | |
| CN111898533B (en) | Gait classification method based on space-time feature fusion |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20201203 Address after: 201210, Zhangjiang Road, Zhangjiang, Pudong New Area, No. 497, Gauss Road, Shanghai Applicant after: SHANGHAI IC R & D CENTER Co.,Ltd. Applicant after: Shanghai IC equipment Material Industry Innovation Center Co.,Ltd. Address before: 201210, Zhangjiang Road, Zhangjiang, Pudong New Area, No. 497, Gauss Road, Shanghai Applicant before: SHANGHAI IC R & D CENTER Co.,Ltd. |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |